Optimize CPDF_CMapParser

In this parser, a vector of CodeRange values are built up during
operations that need to be sync'd with the CPDF_CMap that is being
initialized. In the existing implementation, the vector being built as
a member var for the parser, and copying the values over to the cmap
whenever there is a change. When profiling, this copy is where the code
spends most of its time.

The code has been rewritten to have the parser reference/modify the
instance of the vector in the cmap instead of having its own copy. This
removes all of the copies and significantly speeds things up.

BUG=chromium:881678

Change-Id: Ib8e75962507ca3d3b1ed066fd1faa4fbb7141122
Reviewed-on: https://pdfium-review.googlesource.com/42350
Commit-Queue: Ryan Harrison <rharrison@chromium.org>
Reviewed-by: Henrique Nakashima <hnakashima@chromium.org>

3 files changed, 13 insertions, 7 deletions

diff --git a/core/fpdfapi/font/cpdf_cmap.h b/core/fpdfapi/font/cpdf_cmap.h
index 228c207876..68ad26cfa3 100644
--- a/core/fpdfapi/font/cpdf_cmap.h
+++ b/core/fpdfapi/font/cpdf_cmap.h
@@ -66,8 +66,11 @@ class CPDF_CMap final : public Retainable {
 
   void SetVertical(bool vert) { m_bVertical = vert; }
   void SetCodingScheme(CodingScheme scheme) { m_CodingScheme = scheme; }
-  void SetMixedFourByteLeadingRanges(std::vector<CodeRange> range) {
-    m_MixedFourByteLeadingRanges = range;
+  const std::vector<CodeRange>& GetMixedFourByteLeadingRanges() {
+    return m_MixedFourByteLeadingRanges;
+  }
+  void AppendMixedFourByteLeadingRanges(const CodeRange& range) {
+    m_MixedFourByteLeadingRanges.push_back(range);
   }
 
   int GetCoding() const { return m_Coding; }
diff --git a/core/fpdfapi/font/cpdf_cmapparser.cpp b/core/fpdfapi/font/cpdf_cmapparser.cpp
index a88448c5bd..5bfe17b34f 100644
--- a/core/fpdfapi/font/cpdf_cmapparser.cpp
+++ b/core/fpdfapi/font/cpdf_cmapparser.cpp
@@ -109,14 +109,17 @@ void CPDF_CMapParser::ParseWord(const ByteStringView& word) {
     m_Status = 0;
   } else if (m_Status == 7) {
     if (word == "endcodespacerange") {
-      size_t nSegs = m_CodeRanges.size();
+      auto code_ranges = m_pCMap->GetMixedFourByteLeadingRanges();
+      size_t nSegs = code_ranges.size();
       if (nSegs == 1) {
-        m_pCMap->SetCodingScheme((m_CodeRanges[0].m_CharSize == 2)
+        m_pCMap->SetCodingScheme((code_ranges[0].m_CharSize == 2)
                                      ? CPDF_CMap::TwoBytes
                                      : CPDF_CMap::OneByte);
       } else if (nSegs > 1) {
         m_pCMap->SetCodingScheme(CPDF_CMap::MixedFourBytes);
-        m_pCMap->SetMixedFourByteLeadingRanges(m_CodeRanges);
+        for (auto range : m_PendingRanges)
+          m_pCMap->AppendMixedFourByteLeadingRanges(range);
+        m_PendingRanges.clear();
       }
       m_Status = 0;
     } else {
@@ -126,7 +129,7 @@ void CPDF_CMapParser::ParseWord(const ByteStringView& word) {
       if (m_CodeSeq % 2) {
         CPDF_CMap::CodeRange range;
         if (GetCodeRange(range, m_LastWord.AsStringView(), word))
-          m_CodeRanges.push_back(range);
+          m_PendingRanges.push_back(range);
       }
       m_CodeSeq++;
     }
diff --git a/core/fpdfapi/font/cpdf_cmapparser.h b/core/fpdfapi/font/cpdf_cmapparser.h
index c70d408f90..fae65cfe00 100644
--- a/core/fpdfapi/font/cpdf_cmapparser.h
+++ b/core/fpdfapi/font/cpdf_cmapparser.h
@@ -40,7 +40,7 @@ class CPDF_CMapParser {
   UnownedPtr<CPDF_CMap> const m_pCMap;
   int m_Status;
   int m_CodeSeq;
-  std::vector<CPDF_CMap::CodeRange> m_CodeRanges;
+  std::vector<CPDF_CMap::CodeRange> m_PendingRanges;
   std::vector<CPDF_CMap::CIDRange> m_AdditionalCharcodeToCIDMappings;
   ByteString m_LastWord;
   uint32_t m_CodePoints[4];